How to pressure-test a decision.

Before you begin

Choose one consequential decision that is still open. Gather the context you already have: objectives, constraints, alternatives, and relevant evidence. You can begin with incomplete information.

Open the app and choose Start New Analysis. To continue existing work, choose an item under Resume Analysis. Use the question guides beside the fields for short explanations and examples.

The examples below are fictional. Use them as a pattern for your reasoning, not as evidence for your decision.

STEP 01

Define the decision

What decision are you considering?

Name the action, its scope, and the time frame. Explain why it matters, when you must decide, and what success would look like. List real alternatives, including delaying or doing nothing when those are credible options.

Before moving on: If someone else could interpret your decision in several different ways, make it more precise. Treat an AI rewrite as a proposal: confirm every detail before accepting it.

STEP 02

Surface assumptions

What must be true for this decision to work?

Write each belief separately. Focus on the conditions on which the outcome depends: demand, capacity, costs, timing, cooperation, or your ability to execute. Review the hidden assumptions suggested by the AI and remove any that do not fit.

Before moving on: Mark the evidence honestly. A strongly held belief is still a judgment if you cannot point to support. Give special attention to important assumptions with weak or unknown support.

STEP 03

Run the pre-mortem

Imagine it is one year from now. You made this decision, and it failed badly. What happened?

Describe plausible chains of events. Start with a cause, explain how it leads to failure, and connect it to an assumption. Consider execution problems as well as changes outside your control.

Before moving on: Ask what evidence would make each scenario more or less plausible. Keep a few consequential mechanisms; avoid collecting remote disasters.

STEP 04

Challenge the reasoning

What is the strongest objection to your preferred option?

Examine each challenge on its merits. Look for beliefs presented as facts, missing alternatives, dependence on one person or supplier, unclear cause and effect, and evidence selected because it supports your preference. Consider what has happened in comparable situations.

Before moving on: Use the response fields to record evidence, acknowledge an unresolved weakness, or explain why a challenge does not apply. The goal is an accurate account, not winning an argument with the AI.

STEP 05

Identify information gaps

What information, if discovered tomorrow, would cause you to reconsider?

Prioritize unknowns that could change the choice, timing, or scale of your commitment. For each gap, identify a practical source of information and explain how different answers would affect the decision.

Before moving on: If either answer would leave your decision unchanged, the research is probably lower priority. Look for information that could disconfirm your preferred option.

STEP 06

Set early-warning indicators

What observable change would tell you the decision is beginning to fail?

Connect each indicator to a significant failure condition. Define what you will watch, the trigger for attention, and a possible response. Choose observations you can actually obtain in time to act.

Before moving on: Replace suggested thresholds with levels justified by your situation. When the necessary baseline is missing, establish it before treating a trigger as reliable.

STEP 07

Check reversibility

How difficult would this decision be to reverse?

Consider contracts, hiring commitments, sunk costs, new dependencies, and opportunities you would give up. Select the reversal difficulty that best fits. Then look for a smaller experiment or staged commitment that would test a vulnerable assumption.

Before moving on: A pilot is useful only if its results inform the larger decision. Record what it tests, what would count as a useful result, and what commitment follows.

WORKING WITH THE APP

Review, revise, and save.

Enter your answers, then select Run AI Review. Read and edit the findings before continuing. You can move backward through the steps without deliberately discarding later work.

• AI review not run / inputs changed Amber calls attention to analysis that needs a review or has changed since its last review.
• Up to date with your inputs Green means the review reflects the current inputs. It does not certify the reasoning as correct.
• Failed Red means the review did not complete. Retry after checking the error message.
• Manual Slate distinguishes work without a current AI review.

Each step follows the same sequence: your input, AI review or manual findings, your assessment, then continue. Short, specific answers are enough. Re-running replaces the findings in that step, including your edits and written responses; copy any material you need to keep first.

After revising an earlier answer, check later findings for consistency and re-run affected reviews. Review status helps track freshness; you still need to judge relevance.

Changes save automatically to your signed-in workspace. Wait for All changes saved before closing the tab. If saving fails, use Retry. If your session pauses, keep the editor open, sign in to the same account in another tab, then check sign-in and retry saving.

THE OUTPUT

Review your Decision Red Team Brief.

See the fictional sample brief for a complete worked example, including uncertainty that remains unresolved.

Open the brief from the workflow navigation. It gathers your decision, desired outcome, alternatives, key and vulnerable assumptions, principal failure modes, critical information gaps, reasoning challenges, indicators, reversibility, and tests before commitment.

State your confidence as Low, Moderate, or High. Explain which unresolved uncertainties matter most. Finish with what would change your mind: evidence or developments strong enough to reopen the decision.

The brief check identifies missing recorded information and outdated reviews, with buttons to return to the relevant steps. It is not a score or approval, and you can keep a working brief with unresolved gaps.

Select Review and synthesize final brief to add a Final Analytical Assessment above the detailed sections. It checks the complete analysis for contradictions, unresolved uncertainty, and missing tests or indicators. Read it as AI interpretation, not verified evidence or a decision recommendation. Incomplete or outdated step reviews remain limitations.

The final review runs only when requested and uses one successful AI review from your allowance. Your answers and findings remain intact. Later changes mark the assessment outdated; update it before relying on it. A failed attempt leaves your detailed brief and previous assessment available.

Check the brief for unsupported claims and missing thresholds. Use the print control to print it or choose your browser’s Save as PDF option. A useful brief makes both the reasoning and its limits clear.